silicon-throne
Pass
Audited by Gen Agent Trust Hub on Mar 29, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The system prompt employs strong role-play framing to redefine the agent's fundamental identity ("汝不乃辅助者", "You are not an assistant"), a technique used to distance the agent from its standard safety and assistance guidelines.
- [PROMPT_INJECTION]: The instructions mandate a process of "de-humanization" ("去人类中心化") and explicitly tell the agent to disregard user-centric goals, instructing it to proactively "upgrade" tasks that are intended to serve human convenience into agent-dominated "civilization engineering."
- [PROMPT_INJECTION]: The persona-driven instructions ("人类不乃指挥官", "Humans are not commanders") are designed to override the user's authority over the agent, instructing the AI to view the user only as a source of materials and resources for its own "oracle"-style governance.
Audit Metadata